Currently, these models seem to be useful but produce incorrect code sometimes, right? So coders might need to still exist. And they do they actually work well for whole projects, or just snippets of code (I’m sure increasing project size will be an area of rapid improvement). Also the models are trained on existing code, will the models need extra data to train them?
The analogy would fit if we still needed horses to drive manual cars or to help navigate. Or if cars were an interpolation of horse behavior and we studied horses to improve the operation of cars.
Currently, these models seem to be useful but produce incorrect code sometimes, right? So coders might need to still exist. And they do they actually work well for whole projects, or just snippets of code (I’m sure increasing project size will be an area of rapid improvement). Also the models are trained on existing code, will the models need extra data to train them?
The analogy would fit if we still needed horses to drive manual cars or to help navigate. Or if cars were an interpolation of horse behavior and we studied horses to improve the operation of cars.